def main():
dataset = pbh.getSupervisedDataSet()
network = pbh.loadNetwork()
trainer = BackpropTrainer(network, dataset)
"""
print("Training network until convergence...")
trainer.trainUntilConvergence()
"""
try:
while(True):
startTime = time.time()
error = trainer.train()
pbh.saveNetwork(network)
duration = time.time() - startTime
print("Epoch error: %f after time: %fs" %(error, duration))
except KeyboardInterrupt:
print "User stopped training"
import sys
sys.path.append("./data/")
import time, profile, itertools
import ModifiedTTTBoard as MTT
import minmax
import pybrainHelpers as PBH
network = PBH.loadNetwork()
def neural_network_board_value(board):
net_input = [unicode(elem, "utf-8") for elem in board.board]
val = network.activate(net_input)
return val[0]
def neural_network_minmax(board, depth, alpha=-9999999, beta=9999999):
neural_network_minmax.nodes_visited += 1
if board.isLeafNode() or depth == 0:
return (board.boardScore(), board)
is_max_board = True if board.getActivePlayer() == 1 else False
best_child = None
result = None
child_ranks = [(child, neural_network_board_value(child)) for child in board.generateChildBoards()]
if is_max_board == True:
child_ranks = sorted(child_ranks, key=lambda x: x[1], reverse=True)
result = alpha
for child in child_ranks:
old_result = result
result = max(result, neural_network_minmax(child[0], depth-1, result, beta)[0])
if not result == old_result: